deepcharles/ruptures — reverse-engineered prompt

Build me a Python library for offline change point detection in signals and time series. I want scientists or data analysts to give it a noisy sequence or multivariate signal and get back the moments where the behavior changes, like shifts in level, trend, variance, or other regimes.

Keep the public API simple and consistent. A user should be able to create or load a signal, choose a detection method and model, fit it, predict the breakpoints, and plot the result with the detected changes shown clearly. Include exact and faster approximate approaches, plus models that work for both simple numeric signals and more flexible non parametric cases.

Please make it feel like a real reusable Python package, with tests, documentation, examples, and a small demo that generates a piecewise constant signal, runs a kernel based detector, and displays the true and estimated breakpoints. Keep the code modular so new algorithms and cost models can be added later.